[POC] Build a chat interface to the weave-gitops / flux docs

[foot]

Notes

Prior art:

• https://supabase.com/docs (https://github.com/supabase/supabase/pull/12056/files) - supabase
• https://astro-labs.app/docs (https://github.com/gannonh/gpt3.5-turbo-pgvector) - supabase
• ?

Projects of interest:

• https://github.com/hwchase17/langchain/
• https://github.com/jerryjliu/llama_index
• https://github.com/openai/chatgpt-retrieval-plugin
• https://github.com/mpaepper/content-chatbot
• https://github.com/whitead/paper-qa
• ui: https://github.com/ddiu8081/chatgpt-demo/
• ?

Guides / tutorials:

• https://github.com/openai/openai-cookbook/blob/main/examples/Question_answering_using_embeddings.ipynb
• https://simonwillison.net/2023/Jan/13/semantic-search-answers/
• https://betterprogramming.pub/how-to-build-your-own-custom-chatgpt-with-custom-knowledge-base-4e61ad82427e
• ?

Various links

• https://news.ycombinator.com/item?id=35327204 ways to improve the query-to-vector encoding

Approaches

Supabase

??

Build a static docker image with vectors pre-computed with a simple http api

??

Try and use openai's chatgpt-retrieval-plugin

Pros:

• Would be future proof and compatible with a future chatgpt plugin
• Has upsert etc for adding more docs in the future, e.g. via github-actions in weave-gitops/flux

Cons:

• Untested as an approach
• The vector stores that it uses are unfamiliar